C++: why this simple Scope Guard works?

Question

Every looked at scope guard so far has a guard boolean variable. For example, see this discussion: The simplest and neatest c++11 ScopeGuard

But a simple guard works (gcc 4.9, clang 3.6.0):

template <class C>
struct finally_t : public C {
    finally_t(C&& c): C(c) {}
    ~finally_t() { (*this)(); }
};
template <class C>
static finally_t<C> finally_create(C&& c) {
    return std::forward<C>(c);
}
#define FINCAT_(a, b) a ## b
#define FINCAT(a, b) FINCAT_(a, b)
#define FINALLY(...) auto FINCAT(FINALY_, __LINE__) = \
    finally_create([=](){ __VA_ARGS__ })

int main() {
    int a = 1;
    FINALLY( std::cout << "hello" << a << std::endl ; );
    FINALLY( std::cout << "world" << a << std::endl ; );
    return 0;
}

Why no temporary copies destructed? Is it dangerous to rely on this behavior?

Abbreviated

You're observing the effects of Copy Elision (or Move Elision, in this case). Copy Elision is not guaranteed / mandatory, but usually performed by major compilers even when compiling w/o optimizations. Try gcc's -fno-elide-constructors to see it "break": melpon.org/wandbox/permlink/B73EuYYKGYFMnJtR – Walkup

There is a trick to not rely on copy elision, few changes are required. Here is modification of your version which does not rely on copy elision. – Brom

Answer 1

You're observing the effects of Copy Elision (or Move Elision, in this case). Copy Elision is not guaranteed / mandatory, but usually performed by major compilers even when compiling w/o optimizations. Try gcc's -fno-elide-constructors to see it "break": http://melpon.org/wandbox/permlink/B73EuYYKGYFMnJtR

Cleanser